ai-assisted-coding-governance: Instructions file for Codex

AGENTS.md

Version: 1.33 · Last reviewed: 2026-10-12 · Active client: none (internal repository)

Guidance for AI agents working in this repository — the source repo for the AI-assisted coding governance package. This repo applies its own rules to itself.

⚠️ Mandatory rules

Before any task, read and follow core-rules.md — the non-negotiable base for every task (secrets, data, correctness, licensing, provenance, safe agentic actions, compliance, stop-and-ask, client overrides). Then read the module for your task: writing-rules.md before producing or editing documents and content — including documentation about code, such as READMEs, API references, and runbooks — the module that applies to nearly everything here, since this repository is Markdown; and coding-rules.md before writing, editing, or running code, which covers the PowerShell in scripts/ and testing/harness/. If anything below conflicts with these, the stricter rule wins. Load them in one pass before you start reading this repository's code or content — not after you have picked an approach: core-rules.md, the task module, the active client's profile (none here), and the matching craft companion — writing-patterns.md for content, coding-patterns.md for code — scaled to the blast radius; a typo doesn't earn four files. Load means open the file with your file-reading tool — a link you have not opened has not loaded.

Work the way agent-workflow.md describes (work loop, ask-vs-proceed-vs-object boundary, verification, structured hand-off, bounded iteration and adversarial self-review, economy of effort, subagent delegation in brief), and open agent-delegation.md before you spawn a subagent. Read writing-patterns.md before writing or editing a non-trivial document — most work here — and coding-patterns.md before non-trivial code. Where craft meets safety, safety and correctness win.